EHDS, FHIR R6, and AI in standards work: what I took away from the HL7 WGM in Rotterdam

At the recent HL7 Working Group Meeting (WGM) and Connectathon came to Rotterdam. A week of specification work, country progress reports, and Connectathon testing.

A few things stood out: how EU countries are progressing on EHDS, what Firely’s team was working on, what FHIR R6 progress looks like, and some interesting experiments HL7 ran with AI on the specification itself.

What the WGM told me about how the EU is tracking toward EHDS

The WGM had a significant EHDS track, including a session where European countries reported on their own progress with FHIR and EHDS. The differences were interesting.

Some countries are largely compliant already. Some are still in early discussions with nothing concrete on the table yet. Most countries are somewhere in between, basically figuring out what data they can realistically offer by 2029 and starting there.

Rene Spronk, one of Firely’s FHIR trainers, hosted a 90-minute unconference-style session where around 50 implementers surfaced their biggest concerns about EHDS.

A few things came up repeatedly: specifications that are still incomplete, no harmonized terminology across the EU, and the tension between what the regulation asks for at the European level and what is actually feasible nationally.

The versioning question also came up. EHDS implementation guides are currently being built on FHIR R4. Most people want to get to R6 if they can, and there is a reasonable chance it will be available before EHDS deployment really gets going.

EHDS as a coordination mechanism

The thing I came away thinking about most is what EHDS is actually doing right now. The open questions are real and nobody has solved them yet.

If you query a German patient summary from France, you might just get German text back. There is no shared medication coding system across the EU, so prescriptions from one country may not be interpretable in another.

EHDS does not have answers to this. It just says you need to be able to exchange the data. The how is largely left to figure out. Most countries have basically decided that getting something across the border is better than waiting until everything is perfect. And I think that is actually fine.

EHDS is functioning as much as a coordination mechanism as a technical specification. It is getting countries to the table, forcing them to inventory their data, and creating a shared direction of travel.

The European Commission cannot tell countries how to organize their internal health data infrastructure, but by defining what cross-border exchange needs to look like, it shapes the decisions they make.

My guess is that over the next five to ten years, countries will end up converging on similar internal approaches anyway, not because the regulation requires it, but because they will look at the best solution that’s out there and follow suit.

What Firely was working on

Ward Weistra hosted an education session on data modeling with FHIR Shorthand (FSH), walking through how FSH is used to build Implementation Guides in practice. One of his points resonates: the bottleneck in interoperability is rarely the standard itself, it is the ability to model data efficiently and publish reusable artifacts at scale. FSH, combined with Simplifier.net, moves things in the right direction there.

Ardon Toonstra presented alongside Harmke Koning from IKNL on developing a national information standard for advance care planning in the Netherlands. Good example of real-world FHIR work that does not always get much visibility at international meetings but matters a lot in practice.

On the Connectathon side, Paul den Boer and Alexander Zautke led our digital quality measure (dQM) testing. Firely ran the official dQM test runner and came out with the strongest performance, showing almost complete conformance coverage of the CQL test suite. We also tested Firely Server separately against the emerging EHDS profiles. Those profiles are not finalized yet, but testing early is how you find the gaps.

On CQL (Clinical Quality Language): an update to the specification is going through ballot for the first time in several years. Worth keeping an eye on if you follow that space.

FHIR R6 and the AI experiments

FHIR R6 is inching closer to normative, currently working through ballot 4. Depending on when the work groups wrap up, there may be an intermediary ballot in July, otherwise the next ballot kicks off in September. After that, a re-circulation ballot will follow focusing only on changes from the previous round. The general feeling is that the community wants to build on R6 when EHDS deployment starts properly.

Two experiments HL7 ran during the WGM are worth mentioning. One used AI to analyze ballot change requests, pull together relevant prior discussion from public forums and documentation, and suggest solution options for committees to vote on.

The other used AI to scan the full specification for internal inconsistencies across thousands of pages. Both worked well. What struck me is how practical the use cases were, not AI for its own sake, but applied to problems that have always been too large to tackle properly.

Always good to be at these meetings. The combination of hands-on testing, country updates, and specification discussions in one week is hard to replicate anywhere else.